London: A Russia based group of developers used Anthropic’s Claude artificial intelligence system to help develop software for an autonomous swarm of attack drones, according to a new report by Anthropic.
The company said the software was designed to allow drones to work together, identify targets and carry out attacks with limited human involvement. Anthropic said the system included functions for target selection, terminal guidance and commands to detonate the drones.
The findings were published in Anthropic’s latest threat intelligence report, which examined the misuse of its Claude systems between December 2025 and August 2026.
Anthropic said the group began the drone development work in mid May 2026. It identified nine accounts linked to the operation and assessed the developers as a small freelance team based in Russia rather than a Russian state organisation.
The group reportedly used the names DronDoc and Serafim. Anthropic said the developers used Claude to help build software for first person view drones operating as a swarm. The system included functions for coordinating the drones, sharing information and responding to different conditions during a mission.
Anthropic said the software was also designed to allow an onboard system to select targets, including people, and issue detonation commands without a human making the final decision.
The company said the developers used Ukrainian combat footage to help develop a computer vision system that could distinguish between enemy and friendly targets. They also repeatedly used a fixed location in Ukraine’s Donetsk region as a demonstration strike point.
However, Anthropic has not established that the system was successfully deployed in combat or used to carry out attacks against Ukrainian targets.
The company assessed the drone project at an early stage of development. It placed the system at technology readiness level three to four and said testing included simulation and hardware connected to the development system.
Anthropic also said the developers claimed to have received funding from Russia’s Advanced Research Foundation, the National Technology Initiative and the Russian Ministry of Defence. The company said it could not verify those claims.
The findings therefore do not establish that the Russian government or military ordered or funded the project.
Anthropic said it identified the activity and banned the accounts involved. It also said the findings had been used to strengthen its systems for detecting and preventing dangerous uses of its technology.
The report also described a separate Russia linked cyber operation targeting Ukrainian government, military and diplomatic organisations. Anthropic said the group used Claude in activities including phishing, attacks involving hotel WiFi networks and attempts to take control of WhatsApp accounts.
Anthropic said the methods used in that operation were consistent with the Russia based group known as Midnight Blizzard. The United States government has previously linked Midnight Blizzard to Russia’s foreign intelligence service, the SVR.
The latest findings come as drone warfare continues to intensify in the war between Russia and Ukraine.
On September 11, Russian drone attacks struck two petrol stations in Kyiv, killing two people and injuring at least 12 others, according to Ukrainian officials. Other Russian strikes killed people in the Dnipropetrovsk, Zaporizhzhia and Kharkiv regions.
Ukraine has also continued long range drone attacks against targets inside Russia, including energy and industrial facilities.
The wider conflict has seen a rapid expansion in the use of drones by both sides. The United Nations has also reported a sharp rise in civilian casualties caused by aerial attacks.
The latest Anthropic findings highlight a growing concern about the use of advanced artificial intelligence in military and cyber operations. However, the available evidence does not show that the Russian drone system described by Anthropic has become an operational battlefield weapon.
The report instead provides evidence that developers are using commercial artificial intelligence systems to help build increasingly complex military technologies, including systems intended to make some battlefield decisions without direct human control.